powerd adaptive mode latching

[Nate Lawson]

Stefan Lambrev wrote:
> Hi,
> 
> Oliver Fromme wrote:
>> Andrew Atrens wrote:
>>  > # cvs diff -u acpi_perf.c
>>  > Index: acpi_perf.c
>>  > ===================================================================
>>  > RCS file: /usr/repo/freebsd/src/sys/dev/acpica/acpi_perf.c,v
>>  > retrieving revision 1.26
>>  > diff -u -r1.26 acpi_perf.c
>>  > - --- acpi_perf.c 22 Mar 2007 18:16:40 -0000      1.26
>>  > +++ acpi_perf.c 16 Nov 2007 20:11:02 -0000
>>  > @@ -300,9 +300,9 @@
>>  >                         continue;
>>  >  >                 /* Check for duplicate entries */
>>  > - -               if (count > 0 &&
>>  > - -                   sc->px_states[count - 1].core_freq ==
>>  > - -                       sc->px_states[count].core_freq)
>>  > +               if (count > 0 &&
>>  > +                   CPUFREQ_CMP(sc->px_states[count - 1].core_freq,
>>  > +                               sc->px_states[count].core_freq))
>>  >                         continue;
>>  >  >                 count++;
>>  >  >  > It's so simple that it's kind of silly that it (or something
>> like it)
>>  > hasn't been committed yet...
>>  >  > Anyone out there with a commit bit interested in submitting this ?
>>
>> It seems that nobody is responsible for those things right
>> now (or nobody feels responsible).  My PR bin/115513 is
>> also still open.  I'm using that patch for a long time
>> already and wouldn't want to live without it; it improves
>> interactive behaviour a lot on my notebook with powerd.
>>   
> I really hope someone will commit this before 7.0 release, or it will
> never be part of RELENG_7_0
> and not everyone want to track RELENG_7.
> 
> While still here any idea how to make powerd to not lower cpufreq under
> let's say 1000HZ?
> I want to manually set the allowed minimum, because my laptop is lagging
> too much under certain speeds.
> 
> P.S. Nate Lawson & Bruno Ducrot CCed. They are pointed as tech contacts
> for powerd on the pages that I found :) 

I am not sure this patch should be committed as-is.  It might be better
centralized in the cpufreq mid-layer so that all drivers benefit instead
of just acpi_perf.  If there are frequencies that are too close to each
other (no matter what the source driver), it might be good to eliminate
them.

I'll look into it later today.

-Nate